Artificial Intelligence for an automated Insect Biodiversity Monitoring (AIforIBM)

昆虫和草地由于其高度的多样性、对环境胁迫的敏感性以及对多种生态系统功能和服务的重要性而成为研究的热点。生物多样性探索站和其他地方都记录了昆虫丰度和多样性的大幅减少，土地使用的集约化被认为是全球物种丧失的最重要驱动因素之一。据记载，加强草地管理会减少节肢动物的丰度和多样性，但影响因土地利用组成部分和节肢动物研究组而异。这种变异性的机械的理解是阻碍了由资源限制，因为劳动强度的既定监测方法所施加的昆虫监测的粗糙的空间和时间分辨率。因此，昆虫监测的新方法是必要的，使所需的高分辨率数据，并进一步了解土地利用对昆虫群落的影响。基于照片和人工智能图像识别的数字监控系统是实现这些目标的有前途的技术。目前的项目旨在以极高的时间分辨率监测生物多样性探索站草地地块中的飞虫。我们将使用可用的RGB相机陷阱系统，并开发深度学习算法来量化昆虫丰度，形态组多样性和体型分布。我们将利用生物多样性探索中心现有的大量昆虫数据来训练数据，并对人工智能图像识别的性能进行基准测试。部署一个嵌入式传感器系统与移动的连接，以监测昆虫，我们的目标是进行连续的原位测量在生物多样性勘探站的三个地区的草地地块之前，期间和之后不同的土地管理活动。由于我们的数据的高时间分辨率，我们可以调查土地利用活动如何影响昆虫群落的动态和稳定性，通过量化的参考状态，干扰效果，和恢复阶段直接。我们将区分人口和（再）殖民化过程，以响应土地利用的干扰，利用人口和行为过程在不同的时间尺度上运行，例如，繁殖比繁殖快得多，因此，该项目将推进可用于昆虫监测的技术和土地利用对昆虫群落影响的机械理解。

Insects and grasslands are intensively studied ecological models because of their high diversity, sensitivity to environmental forcing, and importance for multiple ecosystem functions and services. Substantial decreases in insect abundances and diversity have been documented in the Biodiversity Exploratories and elsewhere, and land-use intensification has been identified as one of the most important drivers of global species loss. Intensification of grassland management has been documented to reduce the abundance and diversity of arthropods, but effects vary depending on the land-use component and group of arthropods studies. A mechanistic understanding of this variability is hindered by the coarse spatial and temporal resolution of insect monitoring imposed by resource constraints because of the labor intensity of established monitoring methods. Consequently, new methods for insect monitoring are needed to enable the required high-resolution data and to further our understanding of land-use effects on insect communities. Digital monitoring systems based on photographs and AI-enabled image recognition are promising techniques to enable these. The current project aims to monitor flying insects in the grassland plots of the Biodiversity Exploratories at very high temporal resolution. We will use available RGB-camera trap systems and develop deep learning algorithms to quantify insect abundance, morpho-group diversity, and body size distributions. We will be building on the wealth of existing data on insects from the Biodiversity Exploratories for training data and to benchmark the performance of the AI image recognition. Deploying an embedded sensor system with mobile connectivity to monitor insects, we aim to conduct continuous in-situ measurements in the grassland plots of the three regions of the Biodiversity Exploratories before, during, and after different land-management activities. Because of the high temporal resolution of our data, we can investigate how land-use activities affect the dynamics and stability of the insect communities by quantifying the reference state, disturbance effects, and recovery phases directly. We will differentiate between demographic and (re)colonization processes in response to disturbances by land-use, exploiting that demographic and behavioral processes operate at different time scales, for example recolonization being much faster than reproduction, Thereby, this project will advance the technology available for insect monitoring and the mechanistic understanding of the effects of land use on insect communities.